external disturbances and coupling mechanisms in...
TRANSCRIPT
Proceedings of the ASME 2010 International Design Engineering Technical Conferences &Computers and Information in Engineering Conference
IDETC/CIE 2010August 15-18, 2010, Montreal, Quebec, Canada
DETC2010-29004
EXTERNAL DISTURBANCES AND COUPLING MECHANISMSIN UNDERACTUATED HANDS
Ravi Balasubramanian∗, Joseph T. Belter, and Aaron M. Dollar
Department of Mechanical Engineering
Yale University
New Haven, Connecticut 06520.
{ravi.balasubramanian, joseph.belter, aaron.dollar}@yale.edu
ABSTRACT
With the goal of improving the performance of under-
actuated robotic hands in grasping, we investigate the influ-
ence of the underlying coupling mechanism on the robust-
ness of underactuated hands to external disturbance. This
paper identifies unique behaviors in the hand’s response
as a function of the coupling mechanism and the actua-
tion mode the hand is operated in. Specifically, we show
that in conditions when the actuator position is fixed, hands
with single-acting mechanisms exhibit a bimodal behavior
in contrast to hands with double-acting mechanisms that
exhibit a unimodal behavior. We then present an analysis
of how these behaviors influence grasping capability of the
hand and then discuss implications for underactuated hand
design and operation.
1 INTRODUCTION
There has long been a desire to minimize the number of
actuators in robotic and prosthetic hands due to constraints
on size and mass. This has been accomplished by cou-
pling the motion of joints, with many designs having fewer
actuators than degrees of freedom. Such robotic hands,
termed “underactuated”, also exhibit passive adaptability
between joints and digits; that is, in certain situations, these
hands naturally change posture to adapt to the environment.
While the performance and design of underactuated hands
when driven through internal actuation have been analyzed
in prior work [1–5], the behavior of these hands when they
∗Contact author
are driven by external forces is still not well understood.
This “disturbance response” behavior is the subject of this
paper.
Several clever underactuated hand designs have been
proposed in prior work. While the hands may differ in
the number of fingers and the number of links in each
finger, this paper focuses on the differences between the
hands in terms of the coupling transmission design and the
actuation mode for flexion (curling) and extension (open-
ing) finger motion. There are two primary types of actuat-
ing mechanisms: single-acting and double-acting. Single-
acting mechanisms control only one of either the flexion
or extension of the finger (see Figs. 1a and 1c). This is
achieved using an actuator that can only pull (for example,
a single cable routing) or push (for example, a plunger), and
the reverse motion is achieved using springs. Examples of
robotic hands with single-acting mechanisms include the
SDM [1], Balance Bar [6], and 100G robotic hands [7].
In contrast, double-acting mechanisms control both the
flexion and extension motion of the fingers (see Fig 1d).
This is achieved through different methods, such as four-
bar linkages, double cable routing, or gears. Note that
springs may still be used to ensure mechanism stability and
compliance. Examples of robotic hands with double acting
mechanisms include the Laval Hands [8, 9], SPRING [10],
Southampton [11], Graspar [12], BarrettHand [13], and
Obrero [14] robotic hands. In addition, some double-acting
hands such as the BarrettHand and the Southampton hand
feature clutch mechanisms that engage or disengage the
1 Copyright c© 2010 by ASME
(a) (b) (c)
(d)
Fa
Fa
Fa
(e)
τa
Figure 1. EXAMPLES OF UNDERACTUATED HANDS: (A) SINGLE
ACTING CABLE-DRIVEN SYSTEM, (B) SINGLE-ACTING WHEN COU-
PLING BREAKS DOWN (CABLE SLACK), (C) SINGLE-ACTING WITH
EXTRA DEGREE OF FREEDOM (PLUNGER CAN PUSH, BUT NOT
PULL), (D) DOUBLE-ACTING LINKAGE-DRIVEN SYSTEM, (E) CLUTCH
MECHANISM SYSTEM
joint coupling based on actuation conditions (see Fig. 1e).
The manner in which the underactuated hands reshape
in the presence of external forces is important. External
forces can arise in several situations, including unplanned
collisions, vibration of the base, or changing force from
another finger. These events can occur when the hand
approaches an object or when the hand is already closed
on an object. Forces acting on the phalanxes are trans-
ferred through the mechanism all the way back to the actu-
ator. As a result, the effect of these forces is a function of
not only the coupling scheme in the mechanism, but also
the nature of the actuation, including backdrivability and
control mode (for example, force control or position con-
trol). While the equilibrium of underactuated grasping in
the presence of disturbances has been studied before [15],
the influence of the coupling mechanisms on the hand’s re-
sponse is still not clearly understood. A certain disturbance
force might result in grasp failure in one coupling and ac-
tuation scheme, particularly when are a large number de-
grees of freedom, and in a more stable grasp in another
situation. It is therefore important to analyze how various
configurations behave in response to disturbances and use
θ2
θ1
f1
f2
k1
k2
Proximal link
(length l1)
Distal link
(length l2)
X
Y
Figure 2. TWO-LINK REVOLUTE-REVOLUTE FINGER. THE ACTU-
ATION CAN BE EITHER A SINGLE-ACTING OR DOUBLE-ACTING
MECHANISM (SEE FIG. 1)
this understanding to design the most appropriate coupling
and actuation schemes.
In this paper, we first present a framework for studying
the response of underactuated hands to external disturbance
forces, taking into consideration the kinematics of joint
coupling, external forces, and control modes (section 2).
Then in section 3, we present results from an analysis of
the configuration change of two-link underactuated fingers,
incorporating cable-driven systems and linkage driven sys-
tems, in response to external forces. We also present how
these configuration changes potentially affect a grasp. Fi-
nally, in section 4, we present a discussion of the interesting
behaviors arising from the combination of control modes
and the coupling mechanism and how this analysis can in-
form the design and operation of underactuated hands.
2 FRAMEWORK FOR UNDERACTUATED HAND
ANALYSIS
Our framework for analyzing underactuated hands con-
sists of three components: 1) The static equilibrium equa-
tions that relate contact forces on the phalanxes with joint
torques; 2) The kinematic coupling between the phalanxes
and the actuator; 3) The change in robot configuration due
to the external forces in the presence of joint coupling. As
an example, we consider the flexion and extension behavior
of a two-link revolute-revolute finger with a single actua-
tor (see Fig. 2) .
2.1 Static Equilibrium
The relationship between the contact forces and the net
joint torques in the finger (see Fig. 2) may be expressed as
τ = JTc fe, (1)
2 Copyright c© 2010 by ASME
where fe =
(
f1
f2
)
represents the normal contact force on
the proximal and distal links, τ =
(
τ1
τ2
)
the resulting
torque at the joints, and Jc ∈ R2×2 the Jacobian that maps
between the two spaces. For a two-link mechanism, the
Jacobian Jc can be computed as [2]
Jc =
(
k1 0
k2 + l1 cosθ2 k2
)
, (2)
where k1 represents the proximal force location, k2 the
distal force location, l1 the proximal link length, l2 the dis-
tal link length, θ2 the relative angle between the two links.
While this formulation assumes that the contact point can
slide on the link without friction, friction models can easily
be incorporated into this framework. We present a brief dis-
cussion of the influence of the contact types on this analysis
in section 4.
2.2 Kinematics of the Coupling Mechanism
The kinematics of the coupling mechanism may be ex-
pressed as a first-order differential equation in the mecha-
nism’s configuration θ =
(
θ1
θ2
)
and actuator variable θa.
In cable-driven mechanisms, the actuator variable may be
defined as the angle traveled by the actuator pulley over
which the cable travels, while in linkage driven mecha-
nisms, the actuator variable may be defined as the angle
traveled by the actuating link.
For cable driven mechanisms in current underactuated
hands such as the SDM hand, the kinematics of the cou-
pling mechanism may be expressed as
∆θa = r1∆θ1 + r2∆θ2, (3)
where r1 and r2 represent the pulley radii (assuming unit
radius for the actuation pulley).
For four-bar linkage driven mechanisms in current un-
deractuated hands such as the SARAH hand, the kinematics
of the coupling mechanism may be expressed as
∆θa = ∆θ1 +∆θ2R, (4)
where R represents the transmission ratio of the mecha-
nism. Note that for a four-bar linkage mechanism, the
transmission ratio R is a function of joint configuration θ
and link lengths. In this paper, we consider only small joint
configuration changes from a given configuration θ. Thus,
R is treated as a constant for the instantaneous analysis in
this paper.
A closer analysis of (3) and (4) shows that the kine-
matics of both cable-driven mechanisms and linkage-driven
mechanisms can be expressed as
Ja∆θ = ∆θa, (5)
where Ja =(
a1 a2
)
represents the actuator Jacobian of the
mechanism. For linkage-driven systems, a1 equals 1 and
a2 equals R. For cable-driven systems, a1 equals r1 and a2
equals r2. For both types of systems, the transmission ra-
tio may be defined as R = a2/a1. This common structure
between the coupling mechanisms of existing underactua-
tion hands induces the hands to behave in a similar fash-
ion when driven by the actuator [2]. However, as will be
shown, these mechanisms behave differently in the pres-
ence of disturbance forces.
Some double-acting hands like the BarrettHand [13]
and the Southampton hand [11] use either clutch or brake
mechanisms to enable a multimodal joint coupling. For ex-
ample, the BarrettHand switches the joint coupling depend-
ing on the proximal joint torque value, while the Southamp-
ton hand uses brakes to specify the joint coupling. The joint
coupling for the BarrettHand may be expressed as:
∆θ2 = α∆θ2, if τ1 < τmax
∆θ1 = 0, ∆θ2 > 0, if τ1 > τmax
}
, (6)
where τmax represents the maximum allowed torque at the
proximal joint and α the fixed non-adaptive coupling ra-
tio between the two joints. Note that the BarrettHand is
not backdrivable and is not compliant because of a double-
acting worm gear mechanism. Similarly, the Southamp-
ton hand uses a rigid coupling mechanism and is non-
backdrivable because of the lead-screws that drive the
joints. Thus, these hands are rigid to external disturbances.
We will not explore such hands with such rigid coupling
mechanisms further in this paper.
2.3 Robot Configuration Change Due to External
Force
An external disturbance force fe on the robot can cause
a change in configuration ∆θ. The magnitude and direction
of ∆θ in the joint configuration space depends on factors
such as 1) the joint coupling, 2) the direction, magnitude,
and location of the disturbance force, 3) the hand control
mode, and 4) joint compliance. This paper explores the
change in robot configuration as a function of these factors,
specifically exploring how ∆θ varies with force direction
(tending to either flex or extend the link), magnitude, and
3 Copyright c© 2010 by ASME
point of application. We also look at two different con-
trol modes for the actuator: force control and position con-
trol. To reduce the size of the parameter space, we assume
disturbance forces are applied normal to the links (that is,
frictionless contacts), and do not vary joint stiffness. We
also model external forces as constant, but models where
the external force changes with deflection (such as com-
pliant springy contacts) and/or models including geometric
constraints imposed by external contacts can also be incor-
porated using the same framework (albeit with differing re-
sultant behaviors). The contact forces are intentionally kept
simple in this paper in order to focus on the hand configu-
ration change.
The configuration change ∆θ for an external force fe
can be quantified using a Lagrangian view of the work done
by the external forces and the energy stored in the springs in
the presence of the actuation constraints [16]. Specifically,
we can define the Lagrangian L as
L = ∆Ws +∆Wc +∆Wa, (7)
where Ws represents the work done on the springs, Wc the
work done by the external forces, Wa the work done on the
actuator.
The work done on the spring ∆Ws = −1/2∆θT KJ∆θ
and work done by the external forces ∆Wc = f Te Jc∆θ [17]
are similar in form for all of the underactuated mechanisms
we consider. Here KJ =
(
KJ1 0
0 KJ2
)
represents joint stiff-
ness. The work done on the actuator Wa, however, takes dif-
ferent forms depending on the control mode the mechanism
is driven in (see Table 1). In the force-control mode (that
is, actuator force is controlled to be constant while actuator
position can vary), the work is “real”, while in the position
control mode (that is, actuator position is held fixed and
force can vary), the “virtual” work must equal zero [18]. In
the work equation for the position control mode, p is the
pretension in the actuating mechanism that ensures mecha-
nism stability prior to the application of the external distur-
bance fe.
By taking derivatives of the Lagrangian with respect to
the variables and any Lagrange multipliers, we can derive
the static balance equations [16]. Table 1 presents the static
balance equations for the two-link mechanism, one for each
control mode. Note that these equations predict the instan-
taneous small changes in hand posture ∆θ from a statically
stable configuration as a result of the external force and the
actuation mode.
A closer look at the static equations reveals that the sys-
tem response in the decoupled mode (that is, actuator ap-
plies no load or position constraint, such as the case when
a tendon goes slack) is shaped primarily by the joint stiff-
nesses KJ1 and KJ2. The system response in the force con-
trol mode is shaped by the joint stiffnesses KJ1 and KJ2,
the constant actuation force fa, and pulley radii r1 and r2.
Thus, the system response in the force control mode may
be viewed as a modified version of the system response in
the decoupled mode, since the only difference is work done
on the actuator. In contrast, the system response in the po-
sition control mode is shaped by the joint stiffnesses KJ1
and KJ2 and the pulley radii ratio R (due to the kinematic
constraint created by constant cable length), and the actua-
tor pretension p.
3 RESULTS
By using the SDM hand as an exemplar of a hand with
a single-acting mechanism and the SARAH hand as an ex-
emplar of a hand with a double-acting mechanism, we now
present results from the simulation of the models presented
in section 2.3 to compute hand configuration change for
external loads. We also include an analysis of the potential
impact of such finger deviations on a grasp by analyzing
the kinematics of a two-link mechanism. In the interest of
keeping the parameter space small, we assumed that there
was a force only on the distal joint (that is, f1 = 0), but the
core of the framework used for analysis holds for distur-
bance forces on the proximal link also. Table 2 shows the
other parameters used in the analysis.
3.1 Response Variation As a Function of Contact
Force Location and Magnitude
The models in section 2.3 permit a quantification of
configuration change for different external loads. The re-
sults for the double-acting (such as a linkage-driven mecha-
nism) and the single-acting actuation case (such as a cable-
driven mechanism) for the decoupled and force-control
modes are shown in Figs. 3a and 3b. The results for the
double-acting actuation case with position-control mode is
shown in Fig. 3c. These contour plots show the change
in distal joint configuration ∆θ2 of a two-link mechanism
as a function of the magnitude and direction of contact
force (horizontal axis) and the force contact location (verti-
cal axis). Note that similar contour plots can be derived for
the change in proximal joint configuration ∆θ1.
As expected, Fig. 3a shows that for the decoupled
4 Copyright c© 2010 by ASME
Table 1. EFFECT OF CONTROL MODE ON WORK DONE ON ACTUATOR BY EXTERNAL FORCES
Actuation mode Work done on
actuator
Static equations Example scenario
Force control ∆Wa = fa∆θa1 −K∆θ+ JT
c fe + JTa fa = 0 (8) Maintaining fixed cable ten-
sion in SDM hand.
Position control Virtual work
∆Wa =(λ− p)∆θa = 0 2
Ja∆θ = 0
−K∆θ+ JTc fe + JT
a (λ− p) = 0
(9) 1) Maintaining fixed cable
length in SDM hand; 2) Non-
backdrivability in SARAH
hand.
Decoupled (see
Fig. 1b)
Wa = 0 −K∆θ+ JTc fe = 0 (10) Cable slackening in SDM
hand.
1: fa is the constant actuation force. 2: λ is the tendon force resulting from the coupling constraint.
-0.15
-0.1
-0.05
-0.05
0 0
0.05
0.1
0.05-0.4
-0.2
0
0.2
0.4
Force
location
-0.48
-0.24
0 0.48
-30 -20 -10 0 10 20 30
Disturbance force f2 (N)
0
0.02
0.04
0.06
0.08
0.10
k2 (m)
0.24
0.15
-30 -20 -10 0 10 20 30
Disturbance force f2 (N)
-30 -20 -10 0 10 20 30
Disturbance force f2 (N)
(a) (b) (c)
Figure 3. DISTURBANCE RESPONSE OF A TWO-LINK HAND IN (A) DECOUPLED MODE AND (B) FORCE CONTROL MODE. NOTE THAT (A) AND
(B) APPLY FOR BOTH SINGLE-ACTING AND DOUBLE-ACTING MECHANISMS. (C) DISTURBANCE RESPONSE OF A TWO-LINK HAND IN POSITION
CONTROL MODE WITH A DOUBLE-ACTING MECHANISM. THE CONTOURS SHOW VARIATION IN DISTAL JOINT DEVIATION AS A FUNCTION OF
DISTURBANCE FORCE f2 AND ITS LOCATION k2.
mode, the distal joint closes in (∆θ2 positive) for flex-
ion external forces and opens out for extending external
forces (∆θ2 negative). The system response in force con-
trol mode (see Fig. 3b) is similar to the system response in
the decoupled mode, except that the work done on the ac-
tuator slightly “warps” the contours. For the double-acting
mechanism in position control mode (see Fig. 3c), we see
a critical distal force location k2 at which the mechanism
does not move (∆θ2 = 0) for any external force, indicating
that the mechanism is extremely stiff at that force location.
This location has been termed the equilibrium point for the
mechanism in prior work [2, 19].
The results for the single-acting actuation case (such
as a cable-driven mechanism) in position control mode are
shown in Fig. 4. This plot shows the change in distal-
joint configuration change ∆θ2 of a single-acting two-link
mechanism in position control mode with the joint cou-
pling given by (5). The overall response of the system
is shaped by the joint stiffnesses KJ1 and KJ2, the pulley
radii ratio R (due to the kinematic constraint created by
constant cable length), and actuator pretension. The sys-
tem response is essentially a hybrid between the decou-
5 Copyright c© 2010 by ASME
Table 2. SYSTEM PARAMETERS
Proximal joint stiffness KJ1 1 Nm/rad
Distal joint stiffness KJ2 5 Nm/rad
Proximal and distal link length l1 and l2 0.1 m
Proximal pulley radius r1 0.02 m
Pulley radii ratio R 0.6 m
Proximal joint configuration θ1 π/10 rad
Distal joint configuration θ2 π/3 rad
Cable pretension p 10 N
Cable actuation force fa 10 N
External force on proximal link f1 0 N
-0.028
0.028
0.056
0.056
0.084
0.112
0.14
-0.028
-30 -20 -10 0 10 20 30
Disturbance force f2 (N)
0
0.02
0.04
0.06
0.08
0.10
k2 (m)
Force
location
0
0.112
A B
CDE
Figure 4. DISTURBANCE RESPONSE OF AN SINGLE-ACTING TWO-
LINK HAND IN POSITION CONTROL MODE: VARIATION IN DISTAL
JOINT DEVIATION AS A FUNCTION OF DISTURBANCE FORCE f2
AND ITS LOCATION k2.
pled mode (right of the dashed red line) and double-acting
mechanism in position control mode (left of the dashed
line) seen in Figs. 3a and 3c. The dashed (red) line repre-
sents the combination of force magnitude f2 and location k2
at which the inter-joint coupling breaks down (for example,
the cable going slack in the SDM hand). This dashed line
is a function of the tendon tension (p=10 N in this exam-
ple), and would lie on the f2 = 0 line if the pretension p
was zero.
To the left of the dashed line, we have a ∆θ2 = 0 con-
tour line when f2 = 0, and there is an equilibrium point
when ∆θ2 = 0 for any external force, just as in the case
-30 -20 -10 0 10 20 300
0.04
0.08
Disturbance force f2 (N)
k2 (m)
Force
location
p= −40 p=40
p=0
Figure 5. CONTACT FORCE AND LOCATION COMBINATIONS THAT
NULLIFY THE PRETENSION p IN THE ACTUATOR.
of position control for a double-acting system (see Fig. 3c).
But there exists a ∆θ2 = 0 contour to the right of the dashed
line also, due to the flexion motion of the distal link for
flexion forces in the absence of joint coupling. The transi-
tion from a (coupled) position control mode to a decoupled
mode produces interesting behavior when either k2 or f2 is
small (region ABCDE in Fig. 4). The distal joint changes
from flexion to extension to flexion in a small force range
with force location 0.04 m < k2 < 0.075 m. Note that a
similar contour plot can be drawn for proximal link devia-
tion ∆θ1 also.
Fig. 5 shows the combination of external force on the
distal link and its location that nullifies a pre-existing actu-
ation force (varied here between p =−40 N and p = 40 N).
Note that pretension can be positive or negative for double-
acting systems, but can be either only positive or only neg-
ative for single-acting systems. The contours in this plot
are similar to the dashed red line in Fig. 4.
3.2 Variation of the Equilibrium Point in Position Con-
trol Mode
As noted in Fig. 3c, there is a critical value of force lo-
cation k2, called the equilibrium point, at which the under-
actuated hand does not move (∆θ = 0) in position control
mode. The equilibrium point e is a function of the mech-
anism design parameters such as proximal link length l1,
the pulley radii ratio R, and distal joint angle θ, and has the
form e = l1Rcosθ2/(1−R). Fig. 6 shows that the equilib-
rium point moves distal rapidly as the pulley radii ratio R
increases toward unity (for fixed distal joint angle θ2 = π/3
). This indicates that the equilibrium point could be beyond
the distal link length. In contrast, the equilibrium point
moves proximal on the distal link as the pulley radii ra-
6 Copyright c© 2010 by ASME
-0.448
2-0.168-0.112-0.056
0.056 0.112 0.168
0.392
00.05 0.10 0.15 0.20
1.0
1.5
2.0
Proximal link length l1
Pulley
radii
ratio R
0
0.5
Figure 6. VARIATION OF THE EQUILIBRIUM POINT (CONTOURS) AS
A FUNCTION PROXIMAL LINK LENGTH l1 AND THE PULLEY RADII
RATIO R.
tio R decreases toward unity, indicating that the equilibrium
point is below the distal joint.
3.3 Grasping Behaviors Arising from Hand Configu-
ration Change
Fig. 7 shows the joint configuration change space (for
the two-link finger) split into four regions depending on the
direction of potential fingertip movement in the X-direction
and proximal and distal joint configuration change. The
regions represent four behaviors: 1) Squeezing, 2) Caging,
3) Ejection, and 4) Release. Table 3 presents the potential
effect of these behaviors on a grasp. Note that prior work
in grasping has also explored ejection [2, 20] (albeit in a
different form), caging [21], and hold [22] behaviors.
However, a configuration change of an arbitrary direc-
tion is not possible in underactuated hands. When under-
actuated hands are operated in force control mode, then the
direction of proximal or distal joint motion depends pri-
marily on the direction of external force. The direction of
motion of the proximal joint and distal joint would be iden-
tical, indicating a positive slope in the joint configuration
space. The exact magnitude of the slope would depend on
the relative magnitudes of joint compliance. Thus, the hand
would exhibit release behavior for extending forces on the
distal link that overcome the cable tension and squeeze be-
havior for flexion forces on the distal link (see Fig. 8a).
Distal
Joint
∆θ2
(rad)
Proximal Joint ∆θ1 (rad)
0.1 0.2-0.2 -0.1
-0.2
0.2
Squeeze
Release
Cage
Eject
0
-0.1
0.1
0
Figure 7. GRASPING BEHAVIORS OF A TWO-LINK FINGER AS A
FUNCTION OF JOINT DEFLECTION. THE FINGERTIP HAS DISPLACE-
MENT ALONG THE NEGATIVE X DIRECTION IN THE REGION TO THE
RIGHT OF THE RED SOLID LINE.
Table 3. ROBOT HAND BEHAVIORS AND EFFECT ON A GRASP
Behavior Fingertip and joint
motions
Potential effect on
grasp
Squeeze Negative X-motion;
both joints curl in
Enveloping object
with potentially
multiple contacts.
Caging Negative X-motion;
distal joint curls,
proximal joint opens
Enveloping object
with potential distal
contact. Proximal
contact weakens.
Tends to pull object
inward.
Ejection Positive X-motion;
proximal joint curls,
distal joint opens
Potential contact
with proximal
phalanx, but distal
joint moves away
potentially break-
ing contact. Tends
to push object
outward.
Release Positive X-motion;
both joints open
Both phalanxes po-
tentially break con-
tact.
7 Copyright c© 2010 by ASME
CageEject
Cage Eject
-30 0 30Disturbance force f
2 (N)
SqueezeEject
Cage
Eject0
0.04
0.08
k2 (m)
Force
location
(a) (b)
-30 0 30 -30 0 30
(c)
Release
Squeeze
Squeeze
Release
Cage
Eject
1
2
3
4
5
6
7
(d)
1
2
3,5
4
1 2 3 4 5 1 2 3
4
5 6 7
Double-acting
mechanism Single-acting
mechanism
Proximal Joint ∆θ1 (rad)
Distal
Joint
∆θ2
(rad)
0
Figure 8. RESPONSE BEHAVIOR OF TWO-LINK FINGER IN
A) FORCE CONTROL MODE, B) DOUBLE-ACTING POSITION CON-
TROL MODE, AND C) SINGLE ACTING POSITION CONTROL
MODE TO EXTERNAL DISTURBANCE, D) GRASPING BEHAVIOR OF
THE SINGLE-ACTING AND DOUBLE-ACTING MECHANISMS WHEN
VIEWED IN JOINT CONFIGURATION-CHANGE SPACE. THE NUM-
BERED LOCATIONS CORRESPOND TO A VARIATION IN EXTERNAL
LOAD FROM THE EXTENSION DIRECTION TO THE FLEXION DIREC-
TION INDICATED IN (B) AND (C).
However, we get interesting behaviors when the hand
is operated in position control mode. Figs. 8b and 8c show
the qualitative behavior of single-acting and double-acting
underactuated hands to external disturbances when oper-
ated in position control mode with the parameters in Ta-
ble 2. We notice that double-acting mechanisms exhibit
caging and ejection behaviors only, depending on the dis-
tal force location and magnitude. In contrast, we notice
that the single-acting mechanism exhibits caging, ejection,
and squeezing behaviors, depending on the external force
location and magnitude. Note that these grasping behav-
iors arising from configuration changes are a function of
joint configuration, the coupling mechanism, the number of
degrees of freedom in the finger, and external constraints,
and the system will respond differently as the parameters
change.
4 DISCUSSION
One of the goals of our research is to build underactu-
ated robotic hands that offer near-perfect grasping capabil-
ity in different scenarios. Most previous studies of underac-
tuated hands have explored design issues focused on favor-
able behavior when the hands are actuated internally (force
control mode; see Table 1) [1–5].
Even though we consider only a simple two-link fin-
ger in the absence of external constraints in this paper, a
close analysis of these systems in position control mode re-
veals interesting aspects of their behavior. Note that oper-
ating a robotic hand in position control mode is analogous
to using non-backdriveable transmissions, which is advan-
tageous from a power consumption standpoint. We believe
that outlining the behaviors seen in these examples will of-
fer a better understanding the behavior of underactuated
hands with more degrees of freedom and in the presence
of external constraints.
4.1 Bimodal Response of Single-Acting Underactu-
ated Hands In Position Control Mode
While double-acting mechanisms offer a smooth vari-
ation in response in position control mode (see Fig. 3c),
single-acting mechanisms offer a bimodal response in po-
sition control mode (see Fig. 4). Specifically, single-acting
hands like the SDM hand in position control mode behave
similar to double-acting mechanisms when the cable is taut.
However, a sufficiently large external force in the flexion
direction can cause the cable to go slack, and thus elimi-
nate the coupling between the two joints. In this decoupled
mode, the hand behaves like a passive compliant two-link
mechanism and naturally complies with the external force.
The external load at which the transition from the position
control mode to the decoupled mode occurs is determined
by the pretension in the actuation mechanism (see Figs. 4
and 5).
4.2 Grasping Behaviors In Underactuated Hands
The grasping behaviors of underactuated hands have
been analyzed in several contexts, such as the robustness to
uncertainty in hand position relative to the object [23], the
sliding of the hand on the object [2, 24], inadvertent con-
tact forces [25], and adaptability [3–5]. In section 3.3, we
presented a novel approach to qualifying small hand move-
ments in response to external disturbances in the context
of grasping. Specifically, we use the inward movement of
the fingertip and distal joint flexion to distinguish hand be-
8 Copyright c© 2010 by ASME
haviors such as squeezing and caging which tend to wrap
the fingers around an object. In contrast, behaviors such
as ejection and release tend to unwrap the fingers from the
object. Note that this approach of using fingertip motion
to qualify a grasping motion depends on the instantaneous
joint configuration and the link lengths.
Using this metric over the space of hand configura-
tion changes, we noticed that current underactuated hands
in position control mode switch between different behav-
iors (good and bad) depending on circumstances. Specifi-
cally, hands with double-acting mechanisms produce only
caging and ejection behavior. Furthermore, they exhibit
ejection behavior even for flexion disturbance forces (see
Fig. 8b and 8d). Alternatively, hands with single-acting
mechanisms like the SDM hand produce caging, ejection,
and squeeze behavior (see Fig. 8c and 8d). The squeeze be-
havior occurs when the joint coupling breaks down and is
good for grasping since the hand complies with the exter-
nal load in a favorable direction. Interestingly, hands with
single-acting mechanisms produce an ejection behavior for
small external loads (region ABCDE in Fig. 4), but tran-
sition into squeeze behavior for larger forces. If the eject
behavior for small loads (region ABCDE) is to be elimi-
nated in cable-driven systems, we infer from Figs. 4 and 5
that the pretension must be zero. In such a condition, any
flexion load will cause the mechanism to enter the squeeze
behavior. Note that a zero pretension in the actuation mech-
anism is not always possible, particularly when the hand is
already applying forces to a grasped object. More work
is required to determine how to get advantageous behav-
ior from underactuated hands in position control mode for
arbitrary external loads, particularly in the presence of ex-
ternal constraints and as the number of degrees of freedom
increases.
4.3 Effect of Equilibrium Point on Grasping Behav-
iors
The equilibrium point has been used in previous work
as the point towards which a precision grasp’s contacts can
slide (when internally actuated) for the grasp to become
stable [2]. Thus, underactuated hands have been designed
to locate the equilibrium point within the distal link, with-
out which the object may be ejected. From our analysis
of the response of the fingers to external disturbance (see
Figs. 8b and 8c), we notice that another aspect of the equi-
librium point may be exploited as well. Both the double-
acting and single-acting mechanisms in position control
mode (see Figs. 8b and 8c) transition between different be-
haviors (caging and ejection in this instance) at the equilib-
rium point. Thus, the finger could be designed to place the
equilibrium point so as to maximize the areas of cage and
squeeze behaviors across the mechanism’s joint configura-
tion space. More work is required to provide a unified view
of underactuated grasping in the light on internal actuation
and external disturbances.
It is also interesting that since joint stiffnesses do not
affect the equilibrium point, they can be chosen based on
other factors. For instance, the ratio of the joint stiffnesses
can be set in order to ensure that the proximal joint closes
in on an object at a faster rate than the outer joint when ac-
tuated to ensure maximum grasping ability in the presence
of uncertainty (as is suggested in [26]).
4.4 Modeling of Contact Forces
The external forces that we have used in this paper have
been intentionally kept simple so that the analysis can focus
on the deflections of the mechanism. Specifically, we have
used contact forces that slide on the links [2] and remain
constant as the mechanism changes configuration. How-
ever, note that the choice of contact force model does not
affect the core of the framework used for analysis. Exter-
nal forces that arise from compliant and sticky contacts can
easily be included in the Lagrangian formulation [16].
While our paper has quantified the configuration
changes of the hand as a function of external loads, we
have only shown qualitatively how such deflections may in-
fluence grasps. Once the kinematic constraints arising from
contacts with objects and the environment are included, our
framework for analyzing hand configuration change will in-
form how the contact forces change and influence object
stability. This will then permit us to leverage prior work in
quantifying grasp stability [2, 27].
ACKNOWLEDGMENT
The authors thank Lael Odhner for interesting discus-
sions on Lagrangian formulation of mechanical systems.
REFERENCES
[1] Dollar, A. M., and Howe, R. D., 2006. “A robust com-
pliant grasper via shape deposition manufacturing”.
ASME/IEEE Trans. on Mechatron., 11(2), pp. 154–
161.
[2] Birglen, L., Laliberte, T., and Gosselin, C., 2008. Un-
deractuated Robotic Hands. Springer.
9 Copyright c© 2010 by ASME
[3] Hirose, S., and Umetani, Y., 1978. “The development
of soft gripper for the versatile robot hand”. Mecha-
nism and IVlechine Theory,, 13, pp. 351–359.
[4] Rakic, M., 1989. “Multifingered hand with self-
adaptability”. Robotics and Computer-Integrated
Manufact., 3(2), pp. 269–276.
[5] Rovetta, A., 1981. “On functionality of a new me-
chanical hand”. ASME J. Mechanical Design, 103,
pp. 277–280.
[6] Kamikawa, Y., and Maeno, T., 2008. “Underactuated
five-finger prosthetic hand inspired by grasping force
distribution of humans”. In Proc. IEEE/RSJ Internat.
Conf. on Intell. Robots and Sys., pp. 717–722.
[7] Kaneko, M., Higashimori, M., Takenaka, R., Namiki,
A., and Ishikawa, M., 2003. “The 100G capturing
robot too fast to see”. ASME/IEEE Trans. on Mecha-
tronics, 8(1), pp. 37–44.
[8] Gosselin, C., and Laliberte, T., 1996. Underactuated
mechanical finger with return actuation. US Patent 5
762 390.
[9] Laliberte, T., and Gosselin, C., 2002. Actuation sys-
tem for highly underactuated gripping mechanism.
US Patent 6 505 870.
[10] Carrozza, M. C., Suppo, C., Sebastiani, F., MASSA,
B., Vecchi, F., Lazzarini, R., Cutkosky, M. R., and
Dario, P., 2004. “The spring hand: Development of
a self-adaptive prosthesis for restoring natural grasp-
ing”. Autonomous Robots, pp. 125–141.
[11] Crowder, R. M., Dubey, V. N., Chappell, P. H., and
Whatley, D. R., 1999. “A multi-fingered end effec-
tor for unstructured environments”. In Proc. IEEE In-
ternat. Conf. on Robotics and Automation, pp. 3038–
3043.
[12] Crisman, J. D., Kanojia, C., and Zeid, I., 1996. “Gras-
par: A flexible, easily controllable robotic hand”.
IEEE Robotics and Automation Magazine, pp. 32–38.
[13] Townsend, W. T., 2000. “The barretthand grasper—
programmably flexible part handling and assembly”.
Industrial Robot: An International Journal, 27(3),
pp. 181–188.
[14] Torres-Jara, E., 2005. “Obrero: A platform for sen-
sitive manipulation”. In Proc. IEEE-RAS Internat.
Conf. on Humanoid Robots, pp. 327–332.
[15] Kragten, G. A., Kool, A. C., and Herder, J. L., 2009.
“Ability to hold grasped objects by underactuated
hands: performance prediction and experiments”. In
Proc. of IEEE Internat.Conf. Robotics and Automat.,
pp. 2493–2498.
[16] Inoue, T., and Hirai, S., 2008. Mechanics and Control
of Soft-fingered Manipulation. Springer.
[17] Craig, J. J., 1989. Introduction to Robotics. Addison
Wesley.
[18] Murray, R. M., Li, Z. X., and Sastry, S. S., 1994. A
Mathematical Introduction to Robotic Manipulation.
CRC Press.
[19] Rizk, R., Krut, S., and Dombre, E., 2007. “Grasp-
stability analysis of a two-phalanx isotropic underac-
tuated finger”. In Proc. IEEE/RSJ Internat. Conf. on
Intell. Robots and Sys., pp. 3289–3294.
[20] Begoc, V., Durand, C., Kurt, A., Dombre, E., and
Pierrot, F., 2006. “On the form-closure capability of
robotic underactuated hands”. In Proc. of the Internat.
Conf. on Control, Automation, Robotics, and Vision,
pp. 2011–2018.
[21] Diankov, R., Srinivasa, S., Ferguson, D., and Kuffner,
J., 2008. “Manipulation planning with caging grasps”.
In Proc. of IEEE Internat. Conf. on Humanoid Robots,
pp. 285–292.
[22] Kragten, G. A., and Herder, J. L., 2010. “The abil-
ity of underactuated hands to grasp and hold objects”.
Mechanism and Machine Theory, 45, pp. 408–425.
[23] Dollar, A. M., and Howe, R. D., 2006. “Joint cou-
pling design of underactuated grippers”. In Internat.
Design Engineering Technical Conf and Computers
and Information in Engineering Conf., pp. 903–911.
[24] Kaneko, M., and Hayashi, T., 1993. “Standing-
up characteristic of contact force during selfposture
changing motions”. In Proc. of the IEEE Internat.
Conf. on Robotics and Automation, pp. 202–208.
[25] Dollar, A. M., Jentoft, L., Gao, J. H., and Howe, R. D.,
2010. “Contact sensing and grasping performance
of compliant hands”. Autonomous Robots, 28(1),
pp. 65–75.
[26] Dollar, A. M., and Howe, R. D., 2005. “Towards
grasping in unstructured environments: Grasper
compliance and configuration optimization”. Adv.
Robotics, 19(5), pp. 523–544.
[27] Mason, M. T., 2001. Mechanics of Robotic Manipu-
lation. MIT Press.
10 Copyright c© 2010 by ASME